Code underlying the publication: "Self-Supervised PPG Representation Learning Shows High Inter-Subject Variability"

This repository provides the implementation of a Self-Supervised Learning (SSL) framework for photoplethysmography (PPG) signal representation, as detailed in the paper "Self-Supervised PPG Representation Learning Shows High Inter-Subject Variability." The framework addresses label scarcity in PPG data analysis by utilizing signal reconstruction as a pretext task to learn informative representations, with a focus on applications such as activity recognition. The study highlights that, while SSL improves downstream supervised task performance and enables the use of simpler models, significant inter-subject variability remains a challenge, limiting the model’s generalization capabilities.